There has been speculation that Anthropic hadn’t broadly released Claude Mythos because it didn’t have the compute to serve it, but Anthropic CEO Dario Amodei has said that that isn’t the case.
Speaking on the matter, Amodei was unequivocal: compute isn’t the bottleneck for Mythos. The real reason the model remains restricted is a deliberate, safety-driven decision about how many actors should have access to it — and how quickly that circle should expand.
“Mythos consumes a very tiny fraction of Anthropic’s compute. It’s no issue whatsoever to increase the amount of compute by three X or ten X,” Amodei said. “There’s no contention between tokens for the government and tokens for the private sector.”
That’s a direct rebuttal to the narrative that Mythos has been held back for the same reasons Claude’s consumer plans have been throttled — a squeeze that has been very public and very painful for users. The compute story, Amodei insists, is simply not applicable to Mythos.
The real constraint, he explained, is about risk management. “Where I think they do have legitimate concerns is the idea that if we open up Mythos to too many actors — if we open up Mythos to hundreds of organizations — then there’s this danger.”
But he was quick to add that caution can’t be the only principle at play. “On the other hand, we need to be fair to everyone. We need to fix things as quickly as possible. We’re hoping eventually to get to the point where you have safeguards where we can release Mythos to everyone and block the cyber problems.”
The tension, as Amodei framed it, is between speed and safety — and the clock is running. “The question of how slow versus how fast to go through that gradual release process — it can’t be infinitely slow because, again, the Chinese models are our pacing. They’re six to twelve months behind.”
The statement puts the Mythos situation in a different light from Anthropic’s broader compute troubles. The company has been visibly straining under demand for its consumer-facing models — introducing weekly rate limits, tightening session windows, and cutting off third-party tools — forcing it to strike a compute deal with xAI to relieve the pressure. Mythos, it turns out, exists in a different category entirely: capacity is available, but access is being rationed by design.
The Chinese competition angle is particularly significant. Amodei’s framing — that Chinese frontier models are six to twelve months behind — suggests Anthropic sees a closing window. Moving too slowly on Mythos rollout risks ceding ground; moving too quickly risks enabling the very cyber threats that justify the caution in the first place. It’s a narrow path, and Amodei’s comments suggest Anthropic knows it.